An efficient, robust video fingerprinting system

An efficient, robust system for machine identification of file and stream-based video content is presented. Efficiency is achieved through easily computed features, simple comparisons, and careful selection of robust indices that lead to fast searches. Robustness is achieved by selection of features that reflect the time structure of the content—a measure of how the visual content changes over time, perhaps the quintessential aspect of video. These features, primarily the overall luminance and interframe luminance differences, are unlikely to change as the underlying signal is distorted by typical video processing, both benevolent and otherwise. Feature extraction, indexing, and matching are discussed.